Closed BurntSushi closed 3 months ago
It looks like the buggy CI configuration was introduced here (which is also when I introduced the s390x and powerpc64 targets): https://github.com/BurntSushi/memchr/pull/129
I'm pretty sure I was running the big endian tests locally as part of working on #129, which is why we never had any surprising test failures.
This PR is on crates.io in memchr 2.7.4
.
Resubmitted as #154.
It turns out that #151 broke big endian targets and CI didn't catch it because of a misconfiguration. (Namely,
cross
wasn't being used to run tests even when it was being used to do builds.)I spent about 15 minutes trying to fix this myself, but it quickly became clear I'd need to spend more time writing out the bit arithmetic to get it right.
cc @stepancheg - I'd be happy to take a new PR that fixes the optimization for big endian targets.
Fixes #152